Network and state specificity in connectivity-based predictions of individual behavior

Area of Science:

• Neuroscience
• Cognitive Science
• Brain Imaging

Background:

• Predicting individual behavior from brain functional connectivity (FC) is crucial for understanding brain function.
• Task-based FC data may offer better behavioral prediction than resting-state FC.
• Using circumscribed functional networks enhances prediction interpretability.

Purpose of the Study:

• To investigate how matching functional network priors and task states to behavioral domains impacts predictability of individual performance.
• To assess the predictive power of FC for working memory (WM), social cognition (SOCIAL), and emotion processing (EMO).

Main Methods:

• Utilized Human Connectome Project data for large-scale, out-of-sample predictions.
• Analyzed FC from corresponding and non-corresponding states (WM, SOCIAL, EMO, resting-state) and networks (WM, SOCIAL, EMO, whole-brain).
• Evaluated model fit using root mean squared error and coefficient of determination.

Main Results:

• Overall predictive performance was poor across all domains.
• Whole-brain FC slightly outperformed task-specific networks; task-based FC showed a minor advantage for WM prediction.
• No significant benefits were found from matching network, task state, and performance domains.

Conclusions:

• Multivariate FC patterns, during both task and resting states, contain limited information about individual performance levels.
• Current approaches may need reevaluation to better capture brain mechanisms underlying individual differences in mental abilities.